Onaopepo Adekunle
Adv. Artif. Intell. Mach. Learn., 3 (4):1800-1824
Onaopepo Adekunle : Maastricht University
DOI: https://dx.doi.org/10.54364/AAIML.2023.11104
Article History: Received on: 21-Oct-23, Accepted on: 23-Dec-23, Published on: 30-Dec-23
Corresponding Author: Onaopepo Adekunle
Email: seun.adekunle@maastrichtuniversity.nl
Citation: Onaopepo Adekunle, Arno Riedl, Michel Dumontier (2023). Evidence from Data Science on the relationship between Individual Beliefs/Behaviors Survey during COVID-19 Pandemic and Risk preferences. Adv. Artif. Intell. Mach. Learn., 3 (4 ):1800-1824
The factors that affect individual risk preference have been of central interest to
behavioral economics and psychology. However, how to measure it and its
stability remain a challenge. Dutch law requires financial institutions, including
pension providers, to consider customers’ risk preferences when offering their
services. In this study, we apply data science to investigate whether individual
perceptions of risk and social effects associated with COVID-19 are related to
general risk preferences of individuals, while observing whether the relevant
features that influence risk preferences remain consistent. A supervised machine
learning task over two different target measures of risk preferences (a self
reported risk preference measure from the survey and a lottery based
experimental measure) using three datasets: a main study dataset with
(N = 4, 282) adult survey participants in the Netherlands and two pretest survey
dataset with (N = 314) and (N = 306) respectively. The experimental measure
employs a multiple price list to estimate an average number of safe choices of
participants which is an integer, as such, a regression task using lasso regression
over all of the datasets has been carried out to detect the relationship between
behavior during the pandemic and the experimental risk preference measure. The
self reported risk measure employs a likert scale to determine the risk likelihood
of participants, as such, a classification task using a random forest model over
the different risk tolerance classes on all of the datasets has also been studied.
This assesses the stability of correlating such behavioral features to the revealed
preference approach of eliciting risk preference as compared to self-reported risk
preference from surveys. The hypothesis is that the individual choices selected
regarding COVID-19 survey questions reflect the perceived risk of the pandemic,
which is related to the general risk preferences of the individual. We find that
adherence to social distancing and expected changes in relationships with
colleagues, friends and neighbors are relevant to revealed risk preferences while
stockpiling and social distancing are relevant to self-reported risk preferences.
Additionally, results from correlation analysis of revealed risk preferences are more
consistent and hence, more stable than self-reported risk preference